A Pentagon document shows that the US wants to manipulate public opinion from other countries with the aid created by AI, as well as Russia and China

The United States hopes to use automatic learning (“Machine Learning”) to create and distribute propaganda abroad, in an attempt to “influence the foreign target public” and to “suppress the dissident arguments”, according to a document for Special Operations of the US (SOCOM), analyzed by The Intercept.

The document, which the intercept journalists describe as “a kind of wish list” of the Pentagon on military technology in the near future, reveals new details about a wide range of capabilities that Socom hopes to achieve in the next five to seven years.

These include state-of-the-art surveillance cameras, sensors, guided energy weapons and other gadgets, to help operators to locate and eliminate targets. But among the technologies that the American command wants to procure are also automatic learning software, which can be used in the information war.

In order to strengthen their “advanced technological supplements for military information support operations”, the command is looking for a contractor who can “provide a capability that capitalizes Ai AI or Multi-LLM systems with specialized roles to increase the magnitude of influence operations”.

The so-called “agents of” are artificial intelligence systems that use automatic learning models that operate with minimal human instructions or supervision. These systems can be used together with large linguistic models (LLM), such as Chatgpt, which generates text based on user commands.

The US Special Operations Command believes that techniques could be suitable for managing an autonomous propaganda operation:

“The information environment moves too quickly for the military to get adequately involved and influence an audience on the Internet. A program built to support our goals can allow us to control the narratives and influence the audiences in real time,” notes the document.

What does the “wish list” of the Pentagon regarding the AI ​​capabilities also contain

The document also shows that Socom believes that he needs technology that is very close to China's reported capabilities, with software that scan and absorb large volumes of online discussions to better convince a target population or an individual on any given subject.

Socom states that he specifically wants “automated systems that extract information from the online environment, analyze the situation and respond with messages in accordance with its goals. This technology” should be able to respond to posts, suppress dissent arguments and produce reference materials to support arguments and favorable messages. “

The Pentagon pays special attention to those who could denounce its propaganda efforts.

“This program should also be able to access profiles, networks and systems of individuals or groups trying to counteract or discredit our messages. The ability should use the information obtained to create a more targeted message, capable of influencing that specific individual,” notes the document.

He also shows that Socom anticipates the use of generative systems both to conceive propaganda messages and to simulate how this propaganda will be received once launched in the public space.

Socom's wishes list continues by listing a need for “offensive capabilities” based on “Deepfake” images, first reported by the intercept journalists in 2023.

“The prospect for LLMs to create an infinite flow of refined propaganda with expertise has been greeted with alarm-but in general, in the context in which the United States would be, not the author,” notes The Intercept.

Pentagon is forbidden to deliver propaganda to the American public

The laws that govern the Pentagon, but also its own policies, generally prohibit military propaganda campaigns targeting the American public, but the anonymous nature of the Internet makes it difficult to apply this rule.

Socom's spokesman Dan Lessard said in a statement sent to The Intercept journalists that this program of the Command is targeted “peak capabilities, based on artificial intelligence”.

“All AI capabilities are developed and used under the responsible artificial intelligence of the Defense Department, which ensures responsibility and transparency by requesting human supervision and decisions,” he told The Intercept. “Efforts (…) are aligned with US legislation and politics. These operations do not concern the American public and are designed to support national security goals in the face of more and more complex global challenges,” he said.

Tools such as Chatgpt from Openai or Gemini created by Google have increased in popularity, despite their tendency towards factual errors and unpredictable results. But their ability to immediately generate text on almost any subject, written in the tone required by users, could mark a major leap for online propagandists. These tools give users the opportunity to refine messages for an unlimited number of audiences, without the costs and time required for human work.

The practice has already been largely documented. In May 2024, Openai published a report that revealed attempts by some Iranian, Chinese and Russian actors to use the company's instruments in clandestine influence campaigns, but none has proved particularly successful.

Freedom House warned about “the repressive power of artificial intelligence”

A 2023 report of the Freedom House organization, funded by the Washington State Department, warned of “the repressive power of artificial intelligence”, predicting that “the misinformation campaigns assisted by AI will explode as the evil-installed actors develop new ways to bypass the protection and to exploit the open-source model”.

Warning that “AI generative draws the attention of authoritarian regimes”, the Freedom House report mentioned the potential of use by China and Russia. In the internal use of technology in the US, it referred only in a short section of Ron Desantis and Donald Trump's presidential campaigns, as well as a “Deepfake” video with Joe Biden, manipulated to present the former president making transfobe comments.

Andrew Lohn, former director of emerging technology at the Washington National Security Council, says that the extent to which the public should be worried about an automated propaganda machine, capable of having a global impact, depends on the extent of its use.

William Marcellino, a behavioral science specialist at the Rand Corporation Think Tank told The Intercept that such systems are built. “Regimes such as those in China and Russia are engaged in efforts of malignant influence, supported by AI,” he says, adding that groups affiliated to the Chinese state “have explicitly designed AI scale systems for the war for public opinion.”

“The counteracting these campaigns probably requires answers you have on the scale,” he believes.

US is increasingly worried about foreign governments online

Jessica Brandt, a former director for artificial intelligence and emerging technologies at the Brookings Institution Think Tank, has warned since 2023, in a speech in front of American senators, that “LLMs could increase personalization and, therefore, persuasion of information campaigns.”

In an online ecosystem full of AI-based information campaigns, “it is likely to increase skepticism about the existence of objective truth,” she said at the time.

A 2024 study published in PNAS Nexus magazine found that “linguistic models can generate almost as persuasive texts for the American public as materials from real foreign clandestine propaganda campaigns.”

“It was the propaganda based on Ai, and America has to act,” notes an opinion editorial recently published by The New York Times on Golaxy, a software created by the Chinese company Beijing Thinker, which was initially used for the GO game. The authors of the editorial, Brett Benson, professor of political science at the University of Vanderbilt, and Brett Goldstein, former official of the Washington Defense Department, have penciled a dark image, presenting Golaxy as an emerging leader of influence campaigns aligned to the Chinese state.

Golaxy, they warn, can scan the public content on social networks and can produce personalized propaganda campaigns. “The company claims in private that it can use a new technology to reshape and influence public opinion on behalf of the Chinese Government,” indicates a separate article signed by the NYT national security specialist, Julian Barnes, entitled “China resorts to AI in the Information War.”

Benson and Goldstein claim that a “coordinated response” is required at the governmental, academic and private level in the face of these threats. They describe this response as a defensive one: mapping and counteracting foreign propaganda based on AI.

But this is not the thing suggested by the document from Pentagon, as the information obtained by The Intercept shows.

Photo article: Ritu Jethani / Dreamstime.com.